中国专利.Skill

v1.8.4

通用中国专利挖掘发现与交底书生成全流程:扫描项目文档挖掘专利点、讨论融合、基于脱敏模版生成技术交底书、联网查新、生成后自检含逻辑闭环与公式参数一致性。| Patent mining, disclosure drafting, prior-art search, and consistency self-check.

1· 161·0 current·0 all-time

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for handsomestwei/patent-disclosure-skill.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "中国专利.Skill" (handsomestwei/patent-disclosure-skill) from ClawHub.
Skill page: https://clawhub.ai/handsomestwei/patent-disclosure-skill
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install patent-disclosure-skill

ClawHub CLI

Package manager switcher

npx clawhub@latest install patent-disclosure-skill
Security Scan
Capability signals
Crypto
These labels describe what authority the skill may exercise. They are separate from suspicious or malicious moderation verdicts.
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Benign
high confidence
Purpose & Capability
The skill's name/description align with the included tools (docx/pptx→MD conversion, mermaid rendering, CNIPA EPUB search, md↔docx, iteration logging). One metadata mismatch: registry lists no required binaries, but SKILL.md and tools require Python, pip packages, optional Node.js/mmdc and Playwright/Chromium for full functionality. This is expected for the described features but is a practical dependency gap to be aware of.
Instruction Scope
SKILL.md and prompt files explicitly instruct the agent to read project files, convert Office documents, perform prior-art searches (preferentially via epub.cnipa.gov.cn), run local tools, write timestamped .md/.docx outputs and append iteration logs. These actions are within scope of generating and iterating patent disclosures. Note: the agent will read any files/paths you point it at, so giving it a project root grants broad read access to source/docs.
Install Mechanism
There is no registry-level install spec (instruction-only), but the repository includes code that requires optional installs: pip requirements, Playwright (which downloads Chromium), and Node/npm for mermaid rendering. All installs reference standard package managers and official components (pip, playwright, npm) — no obscure download URLs or shorteners. The Playwright/Chromium download is heavyweight and does network fetches during installation.
Credentials
The skill declares no required environment variables or credentials. Some optional environment flags (PLAYWRIGHT_HEADED, EPUB_WAF_MAX_WAIT_SEC, EPUB_RESULT_HTML, PYTHONUTF8) are used to tune scraping behavior; these are reasonable and proportional to the CNIPA scraping and local rendering tasks. No tokens/keys are requested.
Persistence & Privilege
always:false and no special privileges appear requested. The skill writes files into user-designated output directories (timestamped .md/.docx and a revision log) as part of normal operation. It does not modify other skills or system-wide agent configuration.
Assessment
This skill appears to do what it says: it will read files you point it at, optionally convert Office docs, run local Python/Node tools, and (if used) perform browser-based scraping of the CNIPA public site via Playwright. Before installing or running: 1) review/confirm which local directories you give the skill access to (avoid exposing directories with secrets or unrelated credentials); 2) be prepared to install optional dependencies (pip packages, Playwright/Chromium, Node/npm) — Playwright will download a browser binary; 3) inspect tools/* scripts (they are plain Python) if you have supply-chain concerns; 4) run the skill in a sandbox/VM if you want to isolate network and filesystem effects; 5) note the CNIPA scraper includes automation-fingerprinting tweaks (to improve scraping reliability) — this is technical and intended to interact with a JS-heavy public site, not to exfiltrate data; 6) do not expect the skill to request credentials, and revoke access if you see unexpected credential requests. If you want me to, I can highlight specific lines in any tool file (e.g., cnipa_epub_* or mermaid_render.py) for a closer look.

Like a lobster shell, security has layers — review code before you run it.

latestvk976kmk4eesy2aq0ykbkhcbaqd8585en
161downloads
1stars
4versions
Updated 6d ago
v1.8.4
MIT-0

专利挖掘与交底书生成

本技能覆盖 专利点挖掘查新与差异化交底书生成自检完善 全流程;分步指令在 prompts/,每步执行前 Read 对应文件,与步骤的对照见「Prompt 文件映射」。

环境与约定

  • 语言:默认与用户语种一致;专利与法律术语采用行业常用表述。
  • 图示定稿(Step 7)3.2/3.4 用 fenced mermaid;执行方式、mmdc 安装与降级规则见下表「交底书定稿交付」行及 tools/README.md(正文不要求 ASCII 框图或 PlantUML)。

触发条件

在用户使用以下任一方式时启用本技能:

  • 明确提及:专利挖掘、专利点、技术交底书、交底书、专利交底书、查新、现有技术对比等
  • 斜杠或简短指令:如 /patent-disclosure-skill/patent-disclosure/交底书
  • 迭代模式(按意图识别):当用户意图明显是在已有交底书或上一轮输出上继续工作(如改章节、补实施例、补材料、修正参数/事实、调整表述等),无需用户写出「迭代」等固定词,也不必询问是否进入迭代——Agent 应 Read prompts/iteration_context.md,再 Read prompts/merger.md(侧重新材料、扩展合并)或 prompts/correction_handler.md(侧重纠错、与事实或风格不符),严格按该文件开头的「执行门禁」(优先执行,不可跳过)做完合并或纠正另存为新文件{案件名}_{YYYYMMDDHHmmss}.md 与同名 .docx(与首次定稿同一命名规则,见 disclosure_builder.md §7.3 第 5 点),不覆盖旧稿(除非用户明确要求)。禁止在迭代意图已成立时默认回到 Step 3–4 专利点全文分析(除非用户明确要求重新挖掘专利点)。对话中已出现交底书路径、附件或上文刚交付的草稿时,优先按迭代处理。

工具与数据来源

按任务选用能力;具体工具名称以当前 Agent 环境为准。

若扫描范围内含 Word(.docx)PowerPoint(.pptx),须在 Step 2 纳入阅读前用本仓库 docx_to_md.py / pptx_to_md.py 转为 Markdown;依赖 pip install -r requirements.txt,命令与说明见下表对应行。

常见任务与建议方式

任务建议方式
加载分步指令Read${CLAUDE_SKILL_DIR}/prompts/*.md,见下表
读代码、设计文档、PDF、图片文件读取工具;大仓库先用搜索/语义检索定位再精读
Word(.docx)→ Markdown + 抽取图片(扫描前)Bashpython3 ${CLAUDE_SKILL_DIR}/tools/docx_to_md.py --input {path}.docx --output {dir}/{name}.md;图片默认写入与 .md 同级的 {name}_media/;需 pip install -r requirements.txt(含 mammoth);复杂版式可改由所内导出 PDF/MD 再扫
PowerPoint(.pptx)→ Markdown + 抽取图片(扫描前)Bashpython3 ${CLAUDE_SKILL_DIR}/tools/pptx_to_md.py --input {path}.pptx --output {dir}/{name}.md;默认 {name}_media/;需 pip install -r requirements.txt(含 python-pptx);旧版 .ppt 不支持,请先另存为 .pptx;图表/SmartArt 等若未以图片形状嵌入则可能仅能从备注或另行导出补全
罗列目录、按名找文件目录列举 / 按文件名搜索
联网查新(Step 5)执行前 Read prompts/prior_art_search.md中国专利公布公告:优先 Bash 运行 cnipa_epub_search.py须在生成命令前归纳 2~8 个相关度高的语义块执行时须分多次调用每次仅传一个词块,自行按 pub_number 合并多轮 EPUB_HITS_JSON(勿单次工具调用堆多个 argv,见该 prompt)。一步拉取+解析、不写 HTML 落盘;须 pip install -r tools/requirements-cnipa.txtpython -m playwright install chromiumabstract 规定必用同该 prompt。需整句一次 AND 或保存 HTML 时用 cnipa_epub_crawler.py;异常或无果再 WebSearch
交底书定稿交付(须同时 .md + .docx)3.2 系统框图与 3.4 流程图均用 fenced mermaid不要 ASCII 文字流程图/框图。定稿执行 tools/mermaid_render.py:mermaid 转 PNG(失败块保留围栏)后默认生成同名 .docx;若 Word 失败,按 stderr 提示手动运行 md_to_docx.py。详见 tools/README.md
保存交底书路径写入用户指定路径;未指定时可建议 ./outputs/{案件标识}/凡交付的 .md / .docx 须为 {案件名}_{YYYYMMDDHHmmss}(§7.3 第 5 点,含首次定稿与迭代),勿默认覆盖旧稿;outputs/ 整目录默认由 .gitignore 忽略
迭代对话留档每轮 merger / correction 交付后,在案件目录追加 交底书修订对话记录.mdtools/iteration_dialog_log.py 或等价手工),见 prompts/iteration_context.md

Prompt 文件映射

步骤文件用途
Step 1prompts/intake.md边界与输入问题
Step 2prompts/project_scan.md项目文档扫描;.docx/.pptx 先转换再读(见该文件「Office 文档」节);独立图片目录可跳过
Step 3–4prompts/patent_points_analyzer.md候选专利点、融合与选定
Step 5prompts/prior_art_search.md联网查新与分析要求
Step 6prompts/disclosure_preview.md全文前的摘要预览
Step 7prompts/disclosure_builder.md + prompts/template_reference.md交底书结构、脱敏与图示规范;mermaid 系统框图与流程图范例在 template_reference
Step 8prompts/disclosure_self_check.md内部自检,不写入正文
迭代prompts/iteration_context.md迭代意图、落盘命名、修订对话记录 md(含对话/记录时间)
迭代prompts/merger.md新材料增量合并;文首含门禁;输出 {案件名}_{时间戳}.md/.docx
迭代prompts/correction_handler.md对话纠正;文首含门禁;输出 {案件名}_{时间戳}.md/.docx

主流程(执行顺序)

  1. Read intake.md → 执行 Step 1
  2. Read project_scan.md → 执行 Step 2
  3. Read patent_points_analyzer.md → 执行 Step 3–4
  4. Read prior_art_search.md → 执行 Step 5
  5. Read disclosure_preview.md → 执行 Step 6;用户可跳过
  6. Read disclosure_builder.mdRead template_reference.md → 执行 Step 7(首次交付.md/.docx 亦须 {案件名}_{YYYYMMDDHHmmss},§7.3 第 5 点)
  7. Read disclosure_self_check.md → 内部执行 Step 8,修订后交付

禁止:交底书正文中包含「自检清单」章节;自检仅内部使用。


迭代模式(摘要)

启用方式:根据用户自然语言意图判断(见上文「触发条件」),不要求固定关键词,默认不为「是否迭代」打断用户。

  • 补充材料 / 扩展章节Readiteration_context.mdmerger.md;合并结果另存为带时间戳的 .md/.docx(§7.3 第 5 点);追加 交底书修订对话记录.mditeration_dialog_log.py 或手工);完成后必须输出「合并摘要」留档
  • 指出错误 / 与事实或参数不符Readiteration_context.mdcorrection_handler.md;纠正结果另存为带时间戳的 .md/.docx追加对话记录;完成后必须输出「纠正摘要」留档

主流程 Step 7→8 的 disclosure_self_check.md 仍在新稿定稿路径上内部执行。


Agent 自用工作流检查清单

□ 已按步骤 Read 对应 prompts;Step 2 若目录含 Office,已执行 docx_to_md / pptx_to_md 并读了产出 `.md`
□ 识别到「在已有交底书上修改」类意图时,已 Read `iteration_context.md` 并选用 merger 或 correction_handler(而非从头跑扫描);交付为**新** `{案件名}_{时间戳}.md`/`.docx`,未无故覆盖旧稿
□ 执行 merger / correction_handler 后,已在对话中输出该文件要求的留档摘要(合并摘要 / 纠正摘要);案件目录已追加 **`交底书修订对话记录.md`**(或等价日志)
□ 查新完成且写入 1.1 与区别论述(符合 `prior_art_search.md`:**优先** `tools/cnipa_epub_search.py`,**国知局侧已分多次调用、每轮一词,并已自行合并** `EPUB_HITS_JSON`;**`abstract` 必用且已充分理解后再概括**;异常或无果再 **WebSearch**)
□ 除用户明确跳过外,完成摘要预览
□ 脱敏、mermaid(定稿均已渲染为 PNG)、章节引用符合 template_reference;**已交付 .md 与 .docx**,且**文件名符合 §7.3 第 5 点**(**凡交付均含**时间戳后缀);**正文无**技能/示例仓库类文末脚注
□ 自检在后台完成,正文无自检清单章节

Comments

Loading comments...